ARTICLE

Received 16 Jul 2013 | Accepted 17 Apr 2014 | Published 20 May 2014

Recovery and resilience of tropical forests
after disturbance
Lydia E.S. Cole
1,2
, Shonil A. Bhagwat
1,2,3
& Katherine J. Willis
1,2,4,5
The time taken for forested tropical ecosystems to re-establish post-disturbance is of
widespread interest. Yet to date there has been no comparative study across tropical biomes
to determine rates of forest re-growth, and how they vary through space and time. Here we
present results from a meta-analysis of palaeoecological records that use fossil pollen as a
proxy for vegetation change over the past 20,000 years. A total of 283 forest disturbance and
recovery events, reported in 71 studies, are identied across four tropical regions. Results
indicate that forests in Central America and Africa generally recover faster from past
disturbances than those in South America and Asia, as do forests exposed to natural large
infrequent disturbances compared with post-climatic and human impacts. Results also
demonstrate that increasing frequency of disturbance events at a site through time elevates
recovery rates, indicating a degree of resilience in forests exposed to recurrent past
disturbance.
DOI: 10.1038/ncomms4906 OPEN
1
Department of Zoology, Oxford Long-term Ecology Laboratory, University of Oxford, South Parks Road, Oxford OX1 3PS, UK.
2
Department of Zoology,
Biodiversity Institute, Oxford Martin School, University of Oxford, South Parks Road, Oxford OX1 3PS, UK.
3
Department of Geography, Faculty of Social
Sciences, The Open University, Walton Hall, Milton Keynes MK7 6AA, UK.
4
Department of Biology, University of Bergen, P.O. Box 7803, N-5020 Bergen,
Norway.
5
Royal Botanical Gardens, Kew, Richmond, Surrey TW9 3AB, UK. Correspondence and requests for materials should be addressed to L.E.S.C.
(email: lyd@cantab.net).
NATURE COMMUNICATIONS | 5:3906 | DOI: 10.1038/ncomms4906 | www.nature.com/naturecommunications 1
& 2014 Macmillan Publishers Limited. All rights reserved.
T
ropical forests are recognized globally for the important
ecosystem services that they provide. Knowledge of
recovery rates and responses of tropical forest to past
forms of disturbance may facilitate our understanding of the
capacity of these ecosystems to respond to present and future
events. Although it is generally accepted that if left long enough,
tropical rainforest will recover
1
, there is much debate about the
nature of forest re-growth and the time taken for it.
Re-establishment of forest to approximate pre-disturbance
levels, as reported predominantly from single-site
2,3
and
chronosequence studies
4
, typically requires 20200 years,
depending on the variable measured
57
. In tierra rme forests
of the Amazon basin, for example, levels of species richness
comparable to those found in mature forest are re-established in a
quarter of the time taken for equivalent forest biomass
4
. However,
these studies make no explicit reference to the history of
disturbance in the described location, which could account for
signicant variation in rates of recovery between sites, even
among those from similar abiotic and biotic settings. Through
examining recovery rates from 283 past episodes of disturbance
reported in palaeoecological records, we identied the factors that
affected responses of tropical forest ecosystems to perturbation
through time. We addressed two fundamental questions:
rst, how quickly has forest re-established after deforestation
events in the past?, and, second, what factorsdisturbance type,
geographical location, frequency of disturbance eventsaffect
recovery rates?
To determine trends in reforestation rates of tropical forests,
we conducted a meta-analysis of 71 published fossil pollen
records, which focused on four tropical regions: Asia, Africa, and
Central and South America. After identifying studies through a
literature search using Scopus (www.scopus.com), exploring both
key words, such as tropical forest and fossil pollen, and key
authors performing research in these regions, various selection
criteria were applied. Data sets were included based on their
location (within 23N and 23S), evidence of disturbance events
(and associated recovery) over the past 20,000 years and good
chronological control (
14
C dating); a lack of any of these criteria
resulted in the exclusion of the study. From the selected 71
records, each resulting from a sediment core extracted from a
specic geographical location, a data set of 283 disturbance events
was compiled and analysed, with a quarter of records demon-
strating ve or more of these episodes. Disturbance events are
dened as perturbations that result in a sharp loss of forest pollen
in the palynological record and subsequent gain, and include both
catastrophic disturbances, where relative forest pollen may
decline by 4 90% (FD, see Table 1), to smaller perturbations
where declines of o10% may occur during a period of relative
forest stability or longer-term recovery. Forest recovery is
described as the maximum increase in the percentage of forest
pollen in the displayed pollen sum after a decline, before a
stabilizing point or further decline (F
max
, see Table 1). The focus
of this meta-analysis is on the recovery rate (RR) of forest, as
dened uniquely in each individual study (see Supplementary
Data 1), representing the speed and extent to which this
ecological unit re-grows relative to its pre-disturbance abundance
(not always equivalent to an undisturbed state), with the
assumption that vegetation re-establishment, as inferred through
the proxy of fossil pollen, is indicative of sustained ecosystem
functionality. The type of disturbance at each event was classied
according to information on the published pollen diagrams and/
or in the article text (Table 2). The main variables extracted for
forest RR calculations are presented in Table 1.
This novel approach to analysing long-term trends in tropical
forest re-establishment provides fundamental insights into the
functioning and resilience of tropical forest ecosystems. Results
demonstrate that forests in the Central American tropics respond
more rapidly to disturbance than those in Asia, and that recovery
after large infrequent events is more dynamic compared with re-
growth post-climatic or anthropogenic impacts. In addition, it is
the frequency of past disturbance events that appear to play the
most signicant role in determining the rate of reforestation in
these tropical regions: the greater the rate of disturbance through
time, the faster the recovery. These ndings suggest that tropical
forests, in response to an ecologically-relevant history of exposure
to natural disturbances, are adapting to perturbation through
time, evolving mechanisms that increase their resilience in the
face of environmental change.
Table 1 | Main features and variables extracted from published pollen diagrams for recovery rate calculations.
Variable Description Notation
Independent variables
Disturbance type Factor causing impact on forest vegetation, shown on pollen
diagram or referred to in the text
See Table 2 for categories
Geographical attributes Potential inuencers of forest ecology/disturbance response Location, altitude, latitude, longitude
Standardized rate of disturbance
events (SRD)
Average number of disturbance events in a site per 1,000 years SRD(n/(T
1,pre
T
n,max
))*1,000
Response variable & measurements
Recovery Rate (RR) Rate of increase in forest abundance relative to degree of
disturbance-induced change, that is, % increase in forest pollen
abundance per year in relation to pre-disturbance level
RR(((F
max
F
min
)/(F
pre
F
min
))*100)/T
rec
Forest abundance maximum
pre-disturbance (%)
Percentage of forest pollen at maximum point pre-decline
(that is, baseline forest pollen percentage)
F
pre
Forest abundance minimum at
disturbance (%)
Percentage of forest pollen at minimum point during
disturbance event
F
min
Forest abundance at maximum
recovery (%)
Percentage of forest pollen at point of maximum recovery
(before a stabilizing point or further decline)
F
max
Time period of recovery (years) Time period from maximum reduction to maximum recovery T
rec
Forest abundance decline (%) Percentage decline in forest pollen from F
pre
FD((F
pre
F
min
)*100)/F
pre
RRrecovery rate; SRDstandardized rate of disturbance.
Forest recovery is described as the maximum increase in percentage of forest pollen after a decline, before a stabilizing point or further decline. (Forest abundance is used as a crude descriptor of past
vegetation extent reconstructed from fossil pollen, but is not representative of a quantiable forest area.) (n number of disturbance events).
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4906
2 NATURE COMMUNICATIONS | 5:3906 | DOI: 10.1038/ncomms4906| www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
Results
Rates of recovery. Results demonstrate that the majority of past
disturbances were caused by forest clearing, followed by climatic
changes and large infrequent events (n 166, n 87 and n 13,
respectively). For the majority of disturbances, forest re-growth
stopped before 100% recovery either due to the onset of
another disturbance event or the establishment of a lower forest
proportion in the landscape (approximated through fossil forest
pollen percentages); nevertheless, median forest abundance after
disturbance reached 95.5% of pre-disturbance levels (n 283).
Across disturbance types, recovery rates varied greatly, with
projected times to 95.5% forest recovery from the pre-disturbance
baseline ranging from o10 to 6,846 years (n 283), while the
estimated median time was 210 years (n 283) (Fig. 1) and
average time 503 years (s.d. 768.16, n 283).
Recovery rate versus disturbance category. The response rates of
tropical forest to natural versus anthropogenic disturbances were
not signicantly different, though the greatest range in RR across
categories was between climatic and large infrequent disturbances
(linear mixed-effects model, P value 0.072, n 283; Table 3)
(Fig. 2a). RRs were fastest after large infrequent disturbances (LI)
(median 2.84% relative reforestation per year, n 13) (see
Table 2 for details of RR calculation), and notably slower after
climatic changes (C) and human-induced disturbances (FC) (for
both, medians 0.41% relative reforestation per year, n 87 and
n 166, respectively).
Recovery rate versus geographical location. The broad geo-
graphical region in which the forests reside did signicantly affect
the RR (Figs 2b and 3). In general, forest ecosystems in the
Central American tropics recovered signicantly faster than those
in Asia (linear mixed-effects model, P-value 0.027, n 283;
Table 3), the latter of which recovered the slowest, with median
times for 95.5% recovery of 141 years (n 111) and 415 years
(n 58), respectively. In general, RRs were also signicantly
faster in African forests compared with those in South America
and Asia (linear mixed-effects model, P-value 0.041 and 0.009,
n 283, respectively; Table 3). However, latitude did not appear
to have a signicant effect on RR (linear mixed-effects model,
P-value 0.111, n 283; Table 3), along with the longitudinal
and altitudinal setting of the forest (both excluded from Model I
(Eq. 1) when P-values40.100 were demonstrated in initial
regression analyses, n 283).
Frequency of past disturbance. The standardized rate of
disturbance events, SRD, which was calculated by dividing the
number of disturbances in a site by the approximate time over
which they occurred, giving a rate per 1,000 years (Table 1) allows
investigation of whether frequency of past disturbance con-
tributes to the pattern of RR observed across these tropical forests.
Of all covariates tested, the SRD displayed the most signicant
Table 2 | Disturbance types and indicators (proxies) extracted from published palaeoecological studies.
Disturbance source Disturbance type Proxy
Natural Climate (C)
Precipitation (CP)
Sea-level rise (CS)
Oxygen isotopes, re (low levels, not linked to human presence), magnetic susceptibility, lithology
Rainfall, monsoon strength variation, climate drying (CD)
Sea level
Large infrequent (LI) Hurricane (LI-H), landslide (LI-L), re (LI-F), volcano (volcanic ash) (LI-V)
Human (FC) Burning (B) Microfossil & macrofossil charcoal
Forest clearing (FC) Temporary, predominantly resulting from shifting cultivation (SC), or more permanent, generally
selective clearing, or not described (FC) signied by for example, fruit trees, Poaceae, & disturbance
indicators/secondary forest taxa, for example, Arenga and Macaranga, or magnetic susceptibility
Agriculture (Ag) Agricultural indicators, for example, fruit treesFicus, cropsPoaceae
Unclear (U) Disturbance indicators present but specic type not claried
Disturbance types and proxies (used to identify the former) were extracted either from published pollen diagrams or from the associated text. Abbreviations for the different disturbance types are given
in parentheses. Smaller natural perturbations such as tree falls, though not explicitly dened due to the difculty of identication through fossil pollen analysis, may also contribute to large infrequent (LI)
disturbance events. (See Supplementary Data 1 for the full record of disturbance types referred to here.)
120
100
N
u
m
b
e
r

o
f

d
i
s
t
u
r
b
a
n
c
e

e
v
e
n
t
s
80
60
40
20
0
0

<

5
0
5
0

<

1
0
0
1
0
0

<

5
0
0
5
0
0

<

1
,
0
0
0
1
,
0
0
0

+
Time taken for 95.5% recovery (years)
Figure 1 | Number of disturbance events versus years to 95.5% forest
recovery. The relationship between recorded disturbance events and
numbers of years for 95.5% recovery to pre-disturbance forest abundance
baseline at each event (n 283) is shown. Recovery rates were used to
calculate the time taken to a projected 95.5% recovery, that is, the median
extent of recovery for the data set.
Table 3 | Results of multiple regression analysis using Model I.
Variables Coefcient s.e. P-value
Intercept 1.597 0.252 0.000
Latitude 0.020 0.012 0.111
LocationAfrica 0.856 0.315 0.009*
LocationCentral America 0.794 0.351 0.027*
LocationSouth America 0.252 0.223 0.264
Disturbance categoryFC 0.296 0.187 0.115
Disturbance categoryLI 0.764 0.423 0.072
Disturbance categoryU 0.936 0.346 0.007*
Log(SRD) 0.914 0.082 0.000
w
Output for Model I: log(RR)BLatitude LocationDisturbance Category log(SRD), n 283.
LocationAsia and Disturbance CategoryClimate were set as the reference groups for the
initial model output.
*Subgroups within categorical variables that are signicantly different from the reference set.
wSRD is a highly statistically signicant independent predictor of RR.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4906 ARTICLE
NATURE COMMUNICATIONS | 5:3906 | DOI: 10.1038/ncomms4906 | www.nature.com/naturecommunications 3
& 2014 Macmillan Publishers Limited. All rights reserved.
relationship with RR (linear mixed-effects model, P valueo0.001,
n 283; Table 3) (Fig. 2c), indicating that forests that had
experienced a higher frequency of disturbance events in the past
tended to recover more rapidly after each subsequent perturba-
tion. Further analysis demonstrated that of the four different
categories of disturbance recorded, large infrequent events
occurred at the greatest rate (median 10.444 events per 1,000
years, n 13) and forest clearance at the lowest (median 1.072
events per 1,000 years, n 87) (Fig. 4a; Table 4). In terms of
geographical location, forested sites in the African tropics
D
i
s
t
u
r
b
a
n
c
e

c
a
t
e
g
o
r
y
U
LI
FC
C
Recovery rate (RR)
100 10 1 0
L
o
c
a
t
i
o
n
S
C
Af
A
Recovery rate (RR)
100 10 1 0
Standardised rate of disturbance
(SRD; events per 1,000 years)
10 1 0
R
e
c
o
v
e
r
y

r
a
t
e

(
R
R
)
100
10
1
0
Af Africa
U - Unclear
Key
(a)
LI Natural large infrequent
FC Forest clearance
C Climatic changes
(b)
S South America
C Central America
A - Asia
Figure 2 | Composite gure showing recovery rates (RRs) of different covariates. Panels display the relationship between RR and (a) disturbance
categories, (b) geographical locations and (c) against the standardized rate of disturbance (SRD). (a) Box plot of RR for grouped disturbance categories.
C, climate-related factors (n 87); FC, anthropogenic forest clearance, grouping FC, SC, B, Ag and combinations of these (n 166); LI, natural large
infrequent events (n 13); U, cause of disturbance unclear (n 17). (b) Box plot of RR for different location groups. S, South America (n 85);
C, Central America (n 111); Af, Africa (n 29); A, Asia (n 85). (c) Scatter plot illustrating the relationship between the RR and SRD. Throughout,
RR and SRD are plotted on logarithmic axes to accommodate the variability in these data sets. The vertical line in a and b represents the median RR for the
entire data set, that is, 0.455% relative recovery per year. Shaded areas on a and b represent the interquartile range and the whisker lines the 95%
condence intervals for each category.
Central America
n = 85
Median RR = 141 years
South America
n = 111
Median RR = 325 years
Africa
n = 29
Median RR
= 162 years
Asia
n = 58
Median RR = 415 years
1:149,270,833 N
Figure 3 | Map highlighting the Worlds tropical forest ecosystems and the location of all studies included in this meta-analysis. For each of the
four regions, the number of disturbance events (n) and median time to 95.5% recovery of pre-disturbance baseline (median RR) are shown.
(The WWF Biome of tropical and subtropical moist broadleaf forests is shown in green
40
.)
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4906
4 NATURE COMMUNICATIONS | 5:3906 | DOI: 10.1038/ncomms4906| www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
experienced the least frequent disturbance in the past
(median 0.578 events per 1,000 years, n 29) and Central
American forests the most (median 1.278 events per 1,000
years, n 111) (Fig. 4b; Table 4). Although these patterns have
emerged, none of the locations or disturbance categories proved
signicantly different to each other in their SRD (except for
climatic disturbance and forest clearance; Table 4), suggesting, for
example, that forests in Central America have not experienced
considerably more disturbance events per millennium than the
other tropical regions in the past. Greater and more consistent
sample sizes across groups would help to clarify these differences.
Discussion
Results from this meta-analysis show that tropical forests have
been disturbed by a variety of factors over the Late Glacial and
Holocene periods, and recovery has proceeded at varying rates.
Our analyses indicate that for most tropical forest ecosystems, the
length of time taken to re-establish forest to values c. 95% of
former arboreal abundance takes 4200 years. This nding is
supported by palaeoecological studies independent of this analysis
(lacking inclusion requirements), for example (ref.
8
). Only after
approximately 20% of disturbance events did forest recover
within the 42 years reported as the average for complete recovery
in tropical and boreal forests across the world
9
. Length of time is
an important distinction to make because while there are a
number of studies that suggest that any form of disturbed (often
referred to as secondary) forest has less ecological value than
primary forests
10,11
, few of these have examined forests that have
been recovering for 4200 years and therefore are unlikely to be
comparing like-with-like but rather a forest ecosystem in the early
stages of succession. This relatively extended period required for
recovery is relevant to contemporary forestry management where
periods allocated for forest re-growth post-logging are rarely
ecologically suitable
12
and practices are underpinned by
inadequate and biased knowledge
13
. Incorporating a longer
rotation period into forest management plans, to allow for
enhanced regeneration, may improve forest productivity.
Signicant differences in recovery rates suggest that certain
factors will inuence the speed of tropical forest restoration. The
result that recovery is fastest following natural disturbances, such
as hurricanes, is somewhat intuitive; these are processes that have
been occurring throughout the earths history and considered to
play an important role in ecosystem dynamics
14
. Despite being
labelled as large infrequent disturbances, an ecological term that
illustrates the greater intensity, duration and spatial extent of
these perturbations compared with those that typically affect an
ecosystem
15
, it is also the type of perturbation that has occurred
with the greatest frequency (SRD) in the past (Fig. 4a). This
further supports one of the main conclusions of this study that
these ecosystems recover most rapidly after exposure to natural
disturbances, occurring at frequencies through time of ecological
relevance, that is, events per thousands of years (see for example,
the tens of years and novel forms of disturbance pervading many
of these ecosystems today), which results in their greater ability to
respond dynamically, manifest here in an increasing rate of
recovery, and inferred resilience. Similar ndings arose from a
study investigating recovery rates in overexploited marine
populations
16
; moderate levels of stock depletion over an
extended period, that is, 450 years, led to the adaptation of
populations to the historical shing regime, enhancing recovery
rates and thus promoting resilience. This result lends further
support, from a disparate biome to the fundamental concept of
ecological adaptation, whether through life-history changes
16
or
ecological sorting/habitat ltering
17
, over long-term time scales.
For these tropical forest ecosystems, this result implies that
designing anthropogenic forest disturbances that more closely
mimic natural disturbance regimes will result in a lesser decline in
the abundance of trees in the landscape and may markedly reduce
the time required and resources needed for forest regeneration.
The slow recovery rates recorded following human-induced
burning are of particular concern, given that this is the favoured
mode of conversion to agriculture in tropical forests; it is also
sometimes adopted by conservation managers to enhance
diversity through invoking a particular disturbance regime
18
.
D
i
s
t
u
r
b
a
n
c
e

c
a
t
e
g
o
r
y
U
LI
FC
C
Standardised rate of disturbance (SRD;
events per 1,000 years)
10 1 0
L
o
c
a
t
i
o
n
S
C
Af
A
Standardised rate of disturbance (SRD;
events per 1,000 years)
10 1 0
U - Unclear S South America
C Central America
Af Africa
A - Asia
LI Natural large infrequent
FC Forest clearance
C Climatic changes
Figure 4 | Composite gure showing the standardized rate of disturbance (SRD) of different disturbance categories and locations. Box plots
displaying the relationship between the SRD and (a) different grouped disturbance categories and (b) geographical location groups. (See Fig. 2 legend
for details on notation and sample sizes). The SRD is plotted on a logarithmic scale to accommodate the variability in the data set. The vertical line in
a and b represents the median SRD for the entire data set, that is, 1.072 disturbance events per 1,000 years. Shaded areas on a and b represent the
interquartile range and the whisker lines the 95% condence intervals for each category.
Table 4 | Results of multiple regression analysis using
Model II.
Variables Coefcient s.e. P-value
Intercept 0.213 0.200 0.291
LocationAsia 0.308 0.307 0.320
LocationCentral America 0.296 0.283 0.298
LocationAfrica 0.229 0.414 0.519
Disturbance categoryLI 0.021 0.084 0.804
Disturbance categoryC 0.103 0.032 0.002*
Disturbance categoryU 0.075 0.046 0.106
Output for Model II: log(SRD)BLocation Disturbance Category, n 283. LocationAfrica and
Disturbance CategoryForest Clearance were set as the reference groups for the initial model
output.
*The SRD caused by climatic impacts (C) is signicantly different from that caused by forest
clearance (FC).
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4906 ARTICLE
NATURE COMMUNICATIONS | 5:3906 | DOI: 10.1038/ncomms4906 | www.nature.com/naturecommunications 5
& 2014 Macmillan Publishers Limited. All rights reserved.
The slower recovery rate apparent in Asia and South America
compared with Central America and Africa may be attributable to
current abiotic drivers, such as soil fertility
4
and climate
19
, and
biotic inuences such as vegetation life histories determining
functional group composition
5
, tree species diversity
20
and age
structure
7
; variables inherently linked to the regime of past
disturbance events
4,21
. The species and functional diversity of
these tropical forest communities, the initial state of the forest
pre-disturbance and the relative severity of the disturbance
events
3
, for example the resultant impact on seed availability and
seedling vulnerability
22
, were not explicitly considered in this
study (in the most part due to data limitations), though all may
play a signicant role in shaping the rates and patterns of
recovery observed here
23,24
. Future work, investigating in more
detail the impact of these abiotic and biotic components (via the
development of a universal independent proxy in the case of
disturbance magnitude), would extend understanding of the
causal reasons underlying this studys ndings and provide
further knowledge on the ecology of tropical forests, as well as
information for application in ecosystem management and
restoration. Nonetheless, despite the relatively coarse temporal
and spatial scale of the analyses performed here, ndings provide
many valuable insights for the sustainable use of tropical forest
resources, not least estimates of the time required for the
reforestation of these pivotal ecosystems. For the development of
successful ecosystem service markets, trading in, for example,
carbon credits resulting from effective forest restoration, this
information is sorely lacking and urgently required
25
.
In addition, this study has been able to test for the rst time the
hypothesis that increased past exposure to perturbation (in
particular frequency of events per 1,000 years, Fig. 2c) leads to the
evolution of adaptations that enable forested ecosystems across
the tropics to respond more dynamically to future disturbance. As
well as the abiotic and biotic factors mentioned previously, the
faster rates of forest re-growth in Africa and Central America
may result from more extensive histories of disturbance in those
regions. Past perturbations relevant to the ecological context of
this meta-analysis may include human disturbance resulting from
large-scale forest-based ancient civilizations
2628
and extreme
weather events, for example, droughts in Central Africa
29
and
hurricanes in Central America
30
.
This meta-analysis has demonstrated disturbance-recovery
dynamics over long-term time scales, revealing increasing
recovery rates with higher frequency of disturbance (Fig. 2c).
Such a result runs counter to recent studies modelling ecosystem
resilience to contemporary forms and intensities of disturbance,
namely human induced over short time scales, which, to a great
extent, are far from mimicking natural disturbance regimes
13,31
.
For example, refs 32,33 identify slowing recovery rates with
increasing disturbance as a robust indicator of a loss of resilience
in different systems. It is important that a distinction is drawn
between the types of disturbance experienced in the past versus
today, with the impacts of contemporary regimes, and forest
responses, yet to be fully observed. Accounting for both
hypotheses on resilience may require careful consideration of
both the type of perturbation and the temporal and spatial
context of the ecosystem, which determines its ecological memory
and consequently the disturbance and recovery dynamics. For
understanding resilience, this study has demonstrated how crucial
it is to adopt a long-term view in analysis.
Methods
Chronological calibration. Various procedures were used to date the published
diagrams. Where age-depth models were not present in the published study,
the dating programme, clam
34
, in R35 (ref. 35), was used to construct the
best-tting age-depth model for the reported ages. Similarly, clam was used to
calibrate dates reported in
14
C radiocarbon years. Where data were not available in
the reference for surface dates, a default of 55 Cal. yr BP, with error 1, was used,
in line with conventional dating of modern samples
36
. In the absence of age
uncertainties for a radiocarbon date, an average of the value for dates on either side
was used, or in the case of basal samples, the youngest date reported. If no age
uncertainties were reported, a default of 50 years was used, for example in ref. 37.
IntCal09 and SHCal04 curves were used for Northern and Southern Hemisphere
age-depth model construction and calibrations, respectively, except where the latter
had dates older than 10,522
14
C radiocarbon years, in which case the IntCal09
curve was employed. Calibrated ages are reported as the average of the 95.4% two
sigma (2s) calibrated age range that obtains the greatest relative area under the
probability distribution curve.
Measurements and notation. Table 1 documents the focal features and variables
extracted from pollen diagrams, used in calculations of recovery rates. Details of
the disturbance types, and associated indicators, recorded from published
palaeoecological studies are displayed in Table 2.
Data acquisition and analysis. The response metric, RR, calculated as percentage
forest pollen increase per unit time from the fossil pollen diagrams selected for this
meta-analysis, provides a basic measure of forest recovery. RR is measured as an
average across the entire period of forest pollen increase (representing forest
re-establishment), until the maximum increase in percentage of forest pollen
post-decline is achieved, before a stabilizing point or further decline. It therefore
incorporates any non-linear components of the trajectory, which may result, for
example, from low-level perturbations during the period of recovery; it is, however,
the overall rate at which the forest is responding to each disturbance event that is
the variable of interest here. Datathief III
38
was used to extract RR components
from the published fossil pollen diagrams, to reduce potential bias resulting from
visual interpretation and assist in the calculation of an average rate over the entire
period of forest re-establishment.
The species or functional diversity of the individuals contributing to this summed
forest value was not recorded, in part due to lack of reporting of such information
across selected studies. Thus, the computed RR does not take into account the biotic
character, for example, primary versus secondary taxa, and turnover in the
recovering vegetation, both of which form important components of ecological
resilience and may inuence the speed at which the ecosystem recovers from
perturbation
24
. The focus of this meta-analysis is rather on the RR of forest, as
dened uniquely in each individual study (see Supplementary Data 1 for details of
studies), representing the speed and extent to which this ecological unit re-grows
relative to its pre-disturbance abundance, with the assumption that vegetation re-
establishment, as inferred through the proxy of fossil pollen, is indicative of sustained
ecosystem functionality. Under this criteria, a variety of forest types, ranging from
full canopy cover to some fraction of that (indicated by F
pre
values of individual
events; see Supplementary Data 1), are included in this meta-analysis.
Where emergent variables measured from pollen diagrams demonstrated a
skewed pattern, such as F
max
, and sample sizes were quite different between groups
within variables, for example, disturbance type, non-parametric descriptive
statistics were used to illustrate trends in the data.
The distributions of the response variable (RR) and predictor covariates, for
example SRD, were observed and transformations performed, resulting in
logarithms of RR and SRD being used in the nal analyses. Multiple regression
analyses were run in R
35
, using the nlme package
39
, to investigate the relationship
between RR and the range of recorded variables. Model I (equation 1) represents
the linear mixed-effects model used in the analysis of RR:
Model I log RR Latitude Location Disturbance Category log SRD
1
(See Table 3 for model output.) Site ID was included as a random effect, providing
a unique identier for each fossil pollen record, to obviate potential bias resulting
from pseudoreplication and to enable exploration of overall patterns in the data set
without weighting each effect size by their associated group size (that is, number of
events per pollen record); this was considered an inappropriate transformation in
this meta-analysis given that effect sizes within each group varied greatly. Several
covariates were excluded from the nal multiple regression model (Model 1, Eq. 1),
after they demonstrated an insignicant relationship with the response variable,
log(RR) (linear mixed-effects model, P value40.1, n 283), when considering
their partial slopes: altitude (linear mixed-effects model, P-value 0.625, n 283)
and sequential number of the disturbance event (linear mixed-effects model,
P-value 0.728, n 283), both square-root transformed. Through manipulation of
Model I (Eq. 1), comparisons were made of RR across location and disturbance
category groups. To investigate the emergent signicant relationship between RR
and SRD (linear mixed-effects model, P-valueo0.0001, n 283; Table 3), the
degree of correlation between SRD and the location and disturbance category
groups was tested using the following model (Eq. 2):
Model II log SRD Location Disturbance Category 2
(See Table 4 for model output.) Site ID was included as a random effect (proving a
more important component of this model due to the nature of SRD
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4906
6 NATURE COMMUNICATIONS | 5:3906 | DOI: 10.1038/ncomms4906| www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
measurements). A third model, including log(T
rec
) as a covariate (Model III),
demonstrated a signicant relationship with log(SRD) (linear mixed-effects model,
P-valueo0.01, n 283); however, since T
rec
is used in the calculation of RR, this
nding is not independent of the ndings from Model I. In addition, disturbance
events are assumed to occur randomly through time in each location, not as a
function of the time spent in or level of recovery of the forest ecosystem.
References
1. McNamara, S., Erskine, P. D., Lamb, D., Chantalangsy, L. & Boyle, S. Primary
tree species diversity in secondary fallow forests of Laos. Forest Ecol. Manag.
281, 9399 (2012).
2. Cole, R. J., Litton, C. M., Koontz, M. J. & Loh, R. K. Vegetation recovery 16
Years after Feral Pig removal from a wet Hawaiian forest. Biotropica 44,
463471 (2012).
3. Uhl, C., Jordan, C., Clark, K., Clark, H. & Herrera, R. Ecosystem recovery in
Amazon caatinga forest after cutting, cutting and burning, and bulldozer
clearing treatments. OIKOS 38, 313320 (1982).
4. Saldarriaga, J. G., West, D. C., Tharp, M. L. & Uhl, C. Long-term
chronosequence of forest succession in the Upper Rio Negro of Colombia and
Venezuela. J. Ecol. 76, 938958 (1988).
5. Chazdon, R. L. et al. Rates of change in tree communities of secondary
Neotropical forests following major disturbances. Philos. Trans. R Soc. B Biol.
Sci. 362, 273289 (2007).
6. DOliveira, M. V. N., Alvarado, E. C., Santos, J. C. & Carvalho, J. A. Forest
natural regeneration and biomass production after slash and burn in a
seasonally dry forest in the Southern Brazilian Amazon. Forest Ecol. Manag.
261, 14901498 (2011).
7. Pena-Claros, M. Changes in forest structure and species composition during
secondary forest succession in the Bolivian Amazon. Biotropica 35, 450461
(2003).
8. Bush, M. & Colinvaux, P. A. Tropical forest disturbance: palaeoecological
records from Darien, Panama. Ecology 75, 17611768 (1994).
9. Jones, H. P. & Schmitz, O. J. Rapid recovery of damaged ecosystems. PLoS ONE
4, e5653 (2009).
10. Barlow, J. et al. Quantifying the biodiversity value of tropical primary,
secondary, and plantation forests. PNAS 104, 1855518560 (2007).
11. Gibson, L. et al. Primary forests are irreplaceable for sustaining tropical
biodiversity. Nature 478, 378381 (2011).
12. Asner, G. P. et al. Condition and fate of logged forests in the Brazilian Amazon.
Proc. Natl Acad. Sci. USA 103, 1294712950 (2006).
13. Lindenmayer, D. B. & Laurance, W. F. A history of hubriscautionary lessons
in ecologically sustainable forest management. Biol. Conserv. 151, 1116
(2012).
14. Connell, J. H. Diversity in tropical rain forests and coral reefs. Science 199,
13021310 (1978).
15. Turner, M. G. & Dale, V. H. Comparing large, infrequent disturbances: what
have we learned? Ecosystems 1, 493496 (1998).
16. Neubauer, P., Jensen, O. P., Hutchings, J. A. & Baum, J. K. Resilience and
recovery of overexploited marine populations. Science 340, 347349 (2013).
17. Weiher, E. & Keddy, P. Ecological Assembly Rules: Perspectives, Advances,
Retreats (Cambridge University Press, 2001).
18. Burrows, N. D. Linking re ecology and re management in south-west
Australian forest landscapes. Forest Ecol. Manag. 255, 23942406 (2008).
19. Quesada, C. A. et al. Basin-wide variations in Amazon forest structure and
function are mediated by both soils and climate. Biogeosciences 9, 22032246
(2012).
20. Hector, A. et al. The Sabah Biodiversity Experiment: a long-term test of the role
of tree diversity in restoring tropical forest structure and functioning. Philos.
Trans. R Soc. B Biol. Sci. 366, 33033315 (2011).
21. Muller-Landau, H. C. Carbon cycle: sink in the African jungle. Nature 457,
969970 (2009).
22. Nepstad, D. C., Uhl, C., Pereira, C. A. & da Silva, J. M. C. A comparative study
of tree establishment in abandoned pasture and mature forest of eastern
Amazonia. Oikos 76, 2539 (1996).
23. Chazdon, R. L. Tropical forest recovery: legacies of human impact and natural
disturbances. Perspect. Plant Ecol. Evol. Syst. 6, 5171 (2003).
24. Chapin, III F. S. et al. Biotic control over the functioning of ecosystems. Science
277, 500504 (1997).
25. Palmer, M. A. & Filoso, S. Restoration of ecosystem services for environmental
markets. Science 325, 575576 (2009).
26. Heckenberger, M. J., Russell, J. C., Toney, J. R. & Schmidt, M. J. The legacy of
cultural landscapes in the Brazilian Amazon: Implications for biodiversity.
Philos. Trans. R Soc. B Biol. Sci. 362, 197208 (2007).
27. Mueller, A. D. et al. Recovery of the forest ecosystem in the tropical lowlands of
northern Guatemala after disintegration of classic Maya polities. Geology 38,
523526 (2010).
28. Turner, II B. L. & Sabloff, J. A. Classic period collapse of the Central Maya
Lowlands: insights about human-environment relationships for sustainability.
PNAS 109, 1390813914 (2012).
29. Van Gemerden, B. S., Olff, H., Parren, M. P. E. & Bongers, F. The pristine rain
forest? Remnants of historical human impacts on current tree species
composition and diversity. J. Biogeography. 30, 13811390 (2003).
30. Vandermeer, J. & Granzow De La Cerda, I. Height dynamics of the thinning
canopy of a tropical rain forest: 14 years of succession in a post-hurricane forest
in Nicaragua. Forest Ecol. Manag. 199, 125135 (2004).
31. Barlow, J., Gardner, T. A., Lees, A. C., Parry, L. & Peres, C. A. How pristine are
tropical forests? An ecological perspective on the pre-Columbian human
footprint in Amazonia and implications for contemporary conservation.
Biol. Conserv. 151, 4549 (2012).
32. Dakos, V., Ke, S., Rietkerk, M., van Nes, E. H. & Scheffer, M. Slowing down in
spatially patterned ecosystems at the brink of collapse. Am. Nat. 177,
E153E166 (2011).
33. Veraart, A. J. et al. Recovery rates reect distance to a tipping point in a living
system. Nature 481, 357359 (2012).
34. Blaauw, M. Methods and code for classical age-modelling of radiocarbon
sequences. Quat. Geochronol. 5, 512518 (2010).
35. R Core Team. R: A Language and Environment for Statistical Computing
(R Foundation for Statistical Computing, 2012) http://www.R-project.org.
36. Taylor, R. E. Radiocarbon Dating: An Archaeological Perspective (Academic
Press, 1987).
37. Deevey, E. S. et al. Mayan urbanism: Impact on a tropical karst environment.
Science 206, 298306 (1979).
38. Tummers, B. DataThief III. http://datathief.org/ (2006).
39. Pinheiro, J., Bates, D., DebRoy, S. & Sarkar, D., the R Development Core Team.
NLME: Linear and Nonlinear Mixed Effects Models. R package version 3.1-111
(2013).
40. Olson, D. M. et al. Terrestrial ecoregions of the world: a new map of life on
earth. BioScience 51, 933938 (2001).
Acknowledgements
We thank JL Snaddon, P Long, M Macias-Fauria and S Nogue for comments and
assistance, and NERC, UK for funding.
Author contributions
L.E.S.C., S.A.B. and K.J.W. designed the study; L.E.S.C. collected and analysed the data;
L.E.S.C., S.A.B. and K.J.W. wrote the manuscript.
Additional information
Supplementary Information accompanies this paper at http://www.nature.com/
naturecommunications
Competing nancial interests: The authors declare no competing nancial interests.
Reprints and permission information is available online at http://npg.nature.com/
reprintsandpermissions/
How to cite this article: Cole, L. E. S. et al. Recovery and resilience of tropical forests
after disturbance. Nat. Commun. 5:3906 doi: 10.1038/ncomms4906 (2014).
This work is licensed under a Creative Commons Attribution 4.0
International License. The images or other third party material in this
article are included in the articles Creative Commons license, unless indicated otherwise
in the credit line; if the material is not included under the Creative Commons license,
users will need to obtain permission from the license holder to reproduce the material.
To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms4906 ARTICLE
NATURE COMMUNICATIONS | 5:3906 | DOI: 10.1038/ncomms4906 | www.nature.com/naturecommunications 7
& 2014 Macmillan Publishers Limited. All rights reserved.